This is a large-scale and diverse street-level image dataset containing 25,000 high-resolution images classified into 66 categories, of which 37 categories are labels for specific additional instances. This dataset is 5 times larger than the total amount of fine annotations in the Cityscapes dataset and contains images from all over the world under various conditions, such as different weather, seasons, time of day, etc.

The dataset covers diverse, rich geographic information and details, is suitable for image semantic segmentation and instance segmentation, and can be used to develop advanced algorithms for visual road scene understanding.